Catalyst 9.9 For Linux Released, Still XvBA Lacking

[AdrenalineJunky]

VDPAU > XvBA, we've got benchmarks on this very site to demonstrate as much. If there's one video API AMD will implement in catalyst, it should be VDPAU.

that "benchmark" is of a pre-production video driver vs stable video driver, probably with very early xvba video outputs from mplayer vs more mature vdpau video outputs from mplayer

and the two cards used were not even close in performance levels.

bottom line.... trying to draw any conclusions from that "benchmark" either way would be entirely premature.

[Apopas]

VDPAU > XvBA, we've got benchmarks on this very site to demonstrate as much. If there's one video API AMD will implement in catalyst, it should be VDPAU.

Huh? Where are they?

[bridgman]

I think ethana2 is referring to Gwenole's benchmarks linked off this report :

http://www.phoronix.com/scan.php?page=news_item&px=NzM2OA

[Apopas]

I think ethana2 is referring to Gwenole's benchmarks linked off this report :

http://www.phoronix.com/scan.php?pag...item&px=NzM2OA

Thanks a lot bridgman, but indeed what results can someone accept when the benchamrks are between so different in value hardware?

[bridgman]

I believe the rationale is that both low end and high end GPUs have the same video decode block, so within a GPU generation the model shouldn't make that much difference. I think that is generally correct. The issue with these specific benchmarks, flagged both by Gwenole and others, is that the NVidia chip used had an older decode block which did not fully accelerate VC-1.

[BassKnight]

Finally, with 9.9 Stepmania doesn't crash anymore.

With 9.8 and older versions, the whole X server crashed randomnly when playing Stepmania and the system freezes.

[Kano]

VC1 is fast enough with NV, i just did not think that so many Bluray Discs would use it, but it seems it even more common than H264. The picture quality seems to be even a bit better (but could be the input data too). The main problems are a/v sync problems with mplayer (the only player i could directly play m2ts with accelleration).

[Koorac]

I can't get fglrx 9.9 running at all anymore (and somehow not even the older ones that used to work), I'm stuck with radeonhd.
Install works fine, but no matter what I do X hangs on start-up.
First as usual X crashed when loading amdxmm.so (as with every version of the driver) so I renamed the file, but now hanging on start-up instead of crash.

Nothing I can see in the log file
Xorg.0.log:

http://pastebin.org/17798

xorg.conf:

# /.../
# SaX generated X11 config file
# Created on: 2009-07-24T11:36:28+0200.
#
# Version: 8.1
# Contact: Marcus Schaefer <[email protected]>, 2005
# Contact: SaX-User list <https://lists.berlios.de/mailman/listinfo/sax-users>
#
# Automatically generated by [SaX2] (8.1)
# PLEASE DO NOT EDIT THIS FILE!
#

Section "ServerLayout"
	Identifier     "Layout[all]"
	Screen         "Screen[0]" 0 0
	InputDevice    "Keyboard[0]" "CoreKeyboard"
	InputDevice    "Mouse[1]" "CorePointer"
EndSection

Section "Files"
	InputDevices   "/dev/gpmdata"
	InputDevices   "/dev/input/mice"
	FontPath     "/usr/share/fonts/misc:unscaled"
	FontPath     "/usr/share/fonts/local"
	FontPath     "/usr/share/fonts/75dpi:unscaled"
	FontPath     "/usr/share/fonts/100dpi:unscaled"
	FontPath     "/usr/share/fonts/Type1"
	FontPath     "/usr/share/fonts/URW"
	FontPath     "/usr/share/fonts/Speedo"
	FontPath     "/usr/share/fonts/PEX"
	FontPath     "/usr/share/fonts/cyrillic"
	FontPath     "/usr/share/fonts/latin2/misc:unscaled"
	FontPath     "/usr/share/fonts/latin2/75dpi:unscaled"
	FontPath     "/usr/share/fonts/latin2/100dpi:unscaled"
	FontPath     "/usr/share/fonts/latin2/Type1"
	FontPath     "/usr/share/fonts/latin7/75dpi:unscaled"
	FontPath     "/usr/share/fonts/baekmuk:unscaled"
	FontPath     "/usr/share/fonts/japanese:unscaled"
	FontPath     "/usr/share/fonts/kwintv"
	FontPath     "/usr/share/fonts/truetype"
	FontPath     "/usr/share/fonts/uni:unscaled"
	FontPath     "/usr/share/fonts/CID"
	FontPath     "/usr/share/fonts/ucs/misc:unscaled"
	FontPath     "/usr/share/fonts/ucs/75dpi:unscaled"
	FontPath     "/usr/share/fonts/ucs/100dpi:unscaled"
	FontPath     "/usr/share/fonts/hellas/misc:unscaled"
	FontPath     "/usr/share/fonts/hellas/75dpi:unscaled"
	FontPath     "/usr/share/fonts/hellas/100dpi:unscaled"
	FontPath     "/usr/share/fonts/hellas/Type1"
	FontPath     "/usr/share/fonts/misc/sgi:unscaled"
	FontPath     "/usr/share/fonts/xtest"
	FontPath     "/opt/kde3/share/fonts"
EndSection

Section "Module"
	Load  "dri"
	Load  "dbe"
	Load  "freetype"
	Load  "glx"
EndSection

Section "ServerFlags"
	Option	    "AllowMouseOpenFail" "on"
	Option	    "ZapWarning" "on"
EndSection

Section "InputDevice"
	Identifier  "Keyboard[0]"
	Driver      "kbd"
	Option	    "Protocol" "Standard"
	Option	    "XkbLayout" "de"
	Option	    "XkbModel" "microsoftpro"
	Option	    "XkbRules" "xfree86"
	Option	    "XkbVariant" "nodeadkeys"
EndSection

Section "InputDevice"
	Identifier  "Mouse[1]"
	Driver      "mouse"
	Option	    "Buttons" "10"
	Option	    "Device" "/dev/input/mice"
	Option	    "Name" "Logitech USB-PS/2 Optical Mouse"
	Option	    "Protocol" "explorerps/2"
	Option	    "Vendor" "Sysp"
	Option	    "ZAxisMapping" "4 5"
EndSection

Section "Modes"
	Identifier     "Modes[0]"
	ModeLine     "1280x1024" 107.0 1280 1360 1496 1712 1024 1025 1028 1059
	ModeLine     "1280x1024" 87.8 1280 1328 1360 1440 1024 1027 1034 1053 +hsync -vsync
	ModeLine     "1024x768" 65.3 1024 1080 1184 1344 768 769 772 796
	ModeLine     "1024x768" 56.0 1024 1072 1104 1184 768 771 775 790 +hsync -vsync
	ModeLine     "800x600" 38.9 800 832 912 1024 600 601 604 622
	ModeLine     "800x600" 35.5 800 848 880 960 600 603 607 618 +hsync -vsync
	ModeLine     "768x576" 35.5 768 792 872 976 576 577 580 597
	ModeLine     "768x576" 33.0 768 816 848 928 576 579 583 593 +hsync -vsync
EndSection

Section "Monitor"
	Identifier   "Monitor[0]"
	VendorName   "ENC"
	ModelName    "EIZO S1910"
	UseModes     "Modes[0]"
	DisplaySize  376	301
	HorizSync    30.0 - 64.0
	VertRefresh  43.0 - 61.0
	Option	    "DPMS"
	Option	    "PreferredMode" "1280x1024"
EndSection

Section "Device"
	Identifier  "Device[0]"
	Driver      "fglrx"
	Option	    "VideoOverlay" "On"
	Option	    "VideoOverlaySync" "On"
	Option	    "UseFastTLS" "2"
	BusID       "PCI:1:0:0"
EndSection

Section "Screen"
	Identifier "Screen[0]"
	Device     "Device[0]"
	Monitor    "Monitor[0]"
	DefaultDepth     24
	SubSection "Display"
		Virtual   3840 1200
		Depth     15
		Modes    "1280x1024" "1024x768" "800x600" "768x576"
	EndSubSection
	SubSection "Display"
		Virtual   3840 1200
		Depth     16
		Modes    "1280x1024" "1024x768" "800x600" "768x576"
	EndSubSection
	SubSection "Display"
		Virtual   3840 1200
		Depth     24
		Modes    "1280x1024" "1024x768" "800x600" "768x576"
	EndSubSection
	SubSection "Display"
		Virtual   3840 1200
		Depth     8
		Modes    "1280x1024" "1024x768" "800x600" "768x576"
	EndSubSection
EndSection

Section "DRI"
	Group        "video"
	Mode         0666
EndSection

I even compiled a new 2.6.30 kernel but still no change.
X-server is 1.6.3

[gbeauche]

I believe the rationale is that both low end and high end GPUs have the same video decode block, so within a GPU generation the model shouldn't make that much difference. I think that is generally correct. The issue with these specific benchmarks, flagged both by Gwenole and others, is that the NVidia chip used had an older decode block which did not fully accelerate VC-1.

True, performance is the same on RS780G and M98, among some others. Since the UVD2 is used, that won't change anything. However, if you want to implement video decoding through the GPU/shaders, that will obviously make a difference.

Actually, I am now only testing on RS780 and I did not notice any difference. CPU is still as low as with M98 and my ears are happier...

BTW, the NVIDIA module in an AMD riser card somehow made it run at half speed only. ;-)

[The_Monkey_King]

A day after they had some press event where they showed off a 24 monitor setup running Linux (we weren't there)...

The irony abounds...